您好问个问题，参数里没分类个数

[smallliang]

应该在哪添加啊，我看from_pretrained代码里没找到

[real-brilliant]

不需要分类个数 在'bert.py'的‘MyPro() - get_labels’里，直接把'return [0, 1]'改成你的类别名列表

[li-cheng12]

我也能遇到了类似的问题，我把return[0,1]改成我的类别列表后，报了这个错，RuntimeError: CUDA error: device-side assert triggered。去网上查了下，是说类别的数量必须在0到n_classes之间，但是n_classes（分类个数）不知道在什么地方设置。我的类别数量有4000+个，请问有遇到类似的问题吗

[badbubble]

@licheng-pro
如果你数据中的label已经转成了label id:

def get_labels(self):
    return [str(i) for i in range(n_classes)]

[li-cheng12]

@ETCartman 没有转成label id,

并且当我换成cpu后，报了这个错。RuntimeError: Assertion `cur_target >= 0 && cur_target < n_classes' failed. at /pytorch/aten/src/THNN/generic/ClassNLLCriterion.c:93

[badbubble]

@licheng-pro 需要return是lable id，而不是label!
try this:

def get_labels(self, label_path):
    fo = open(label_path, 'r', encoding='utf-8')
    lines = fo.readlines()
    label_to_id ={}
    for i, line in enumerate(lines):
        label_to_id[line.strip()] = i
    print(label_to_id)
    fo.close()
    return [str(i) for i in range(len(label_to_id))]

or just:

def get_labels(self, label_path):
    fo = open(label_path, 'r', encoding='utf-8')
    lines = fo.readlines()
    fo.close()
    return [str(i) for i in range(len(lines))]

[li-cheng12]

@ETCartman
convert_examples_to_features方法里的这个地方才是把label变成label id吧，getlabels方法返回的应该就是label吧?

[badbubble]

@licheng-pro 刚去看了下代码确实是:stuck_out_tongue:，不过我做2k个类别没有任何问题。
RuntimeError: Assertion `cur_target >= 0 && cur_target < n_classes' failed. at /pytorch/aten/src/THNN/generic/ClassNLLCriterion.c:93
我个人感觉这个错误肯定是label出了问题，比如label_path中的数据有重复, 你可以
return list(set(label_list))
试一下

[li-cheng12]

我已经解决了，多谢了，老哥 @ETCartman

[liuyijiang1994]

不过现在好像在

model = BertForSequenceClassification.from_pretrained(args.bert_model,
                                                          cache_dir=PYTORCH_PRETRAINED_BERT_CACHE / 'distributed_{}'.format(
                                                              args.local_rank))

这一行会报如下的错误：

File "/root/anaconda3/envs/liu37/lib/python3.7/site-packages/pytorch_pretrained_bert/modeling.py", line 581, in from_pretrained
   model = cls(config, *inputs, **kwargs)
TypeError: __init__() missing 1 required positional argument: 'num_labels'

[Zhaohaoran1997]

不过现在好像在

model = BertForSequenceClassification.from_pretrained(args.bert_model,
                                                          cache_dir=PYTORCH_PRETRAINED_BERT_CACHE / 'distributed_{}'.format(
                                                              args.local_rank))

这一行会报如下的错误：

File "/root/anaconda3/envs/liu37/lib/python3.7/site-packages/pytorch_pretrained_bert/modeling.py", line 581, in from_pretrained
   model = cls(config, *inputs, **kwargs)
TypeError: __init__() missing 1 required positional argument: 'num_labels'

我也是这个问题，请问你解决了吗？

[liuyijiang1994]

@Zhaohaoran1997 在最后加上这个参数就可以

model = BertForSequenceClassification.from_pretrained(args.bert_model,
                                                          cache_dir=PYTORCH_PRETRAINED_BERT_CACHE / 'distributed_{}'.format(
                                                              args.local_rank), num_labels=len(label_list))

也可以参考我fork的代码:https://github.com/liuyijiang1994/bert_senta

[Zhaohaoran1997]

@Zhaohaoran1997 在最后加上这个参数就可以

model = BertForSequenceClassification.from_pretrained(args.bert_model,
                                                          cache_dir=PYTORCH_PRETRAINED_BERT_CACHE / 'distributed_{}'.format(
                                                              args.local_rank), num_labels=len(label_list))

也可以参考我fork的代码:https://github.com/liuyijiang1994/bert_senta

我改了num_labels之后程序被kill了，请问是我写得有bug还是机器性能不足？

04/22/2019 15:58:31 - INFO - pytorch_pretrained_bert.modeling - Weights of BertForSequenceClassification not initialized from pretrained model: ['classifier.bias', 'classifier.weight'] 04/22/2019 15:58:31 - INFO - pytorch_pretrained_bert.modeling - Weights from pretrained model not used in BertForSequenceClassification: ['cls.predictions.bias', 'cls.predictions.transform.dense.weight', 'cls.predictions.transform.dense.bias', 'cls.predictions.decoder.weight', 'cls.seq_relationship.weight', 'cls.seq_relationship.bias', 'cls.predictions.transform.LayerNorm.weight', 'cls.predictions.transform.LayerNorm.bias'] 04/22/2019 15:58:34 - INFO - __main__ - ***** Running training ***** 04/22/2019 15:58:34 - INFO - __main__ - Num examples = 43425 04/22/2019 15:58:34 - INFO - __main__ - Batch size = 128 04/22/2019 15:58:34 - INFO - __main__ - Num steps = 3392 Epoch: 0%| | 0/10 [00:00<?, ?it/s]已杀死ion: 0%| | 0/340 [00:00<?, ?it/s]

[liuyijiang1994]

@Zhaohaoran1997 看上去运行的时候已经是正常的了 也许是性能的问题